The question is not really “ChatGPT or Claude”. It is which kind of account, and what your team does with it. The same product can be a sensible business tool or a privacy problem depending on how it is set up.

The account matters more than the brand

Both OpenAI and Anthropic now draw a clear line between personal and business plans.

OpenAI says it does not use data from ChatGPT Business, ChatGPT Enterprise or its API to train its models by default, and that the business owns its inputs and outputs. Anthropic says the same of Claude for Work (its Team and Enterprise plans) and its API: those are explicitly excluded from the consumer terms that let it train on chats.

Personal accounts are different. On free and personal ChatGPT plans, conversations can be used for training unless the person switches it off. On Claude Free, Pro and Max, each person chooses whether their chats are used for training, and if they allow it Anthropic keeps that data for up to five years.

So the first safety decision is simple: client information only ever goes into a business plan, on a company login, never into someone’s personal account.

What Australia’s regulator says

The Office of the Australian Information Commissioner has published guidance on using commercially available AI products. Its plainest recommendation is that organisations should not enter personal information, and particularly sensitive information, into publicly available AI chatbots.

It also says the Privacy Act applies to any use of AI involving personal information, and that you should do some due diligence before choosing a product: what it does with your data, who can access it, and how a person stays in the loop. I am not a lawyer and this is not legal advice, but the direction is clear enough to act on.

What I set up for clients

When a business asks me whether AI is safe for client work, this is what “yes” looks like in practice:

  • One approved tool on a business plan. Not five people on five free accounts.
  • Company logins only, so access is switched off in one place when someone leaves.
  • AI connected to your own Microsoft or Google tenant, so files stay where they already live and the AI can only see what that person could already see.
  • A two-page policy that says which accounts, what may go in, and what never does.
  • A person approves anything that goes out. AI drafts the email or the letter; someone reads it before it reaches a client.

What still never goes in

Even on a business plan, I keep a short “never” list: passwords and keys, full identity documents, and anything a client has asked you not to share. In health practices and law firms the bar is higher again, and the work is designed so the AI handles the admin around the record rather than the most sensitive content itself, and uses de-identified data wherever the task allows.

The honest summary

AI on a properly set up business account is roughly as safe as the email and file storage you already trust with client information. AI on a staff member’s free account is a leak waiting to be noticed. The difference is a few hours of setup, which is why it is the first thing I do in almost every business I work with.